Artificial intelligence has shown remarkable promise in medicine, but one of its biggest limitations remains surprisingly fundamental: it can't reliably generate new biological discoveries without first understanding the language of cells.

A new study led by Houston Methodist biomedical engineer Stephen Wong, PhD, offers a potential solution.

Published in Signal Transduction and Targeted Therapy, the research introduces iS2C2, a new "co-intelligent" platform that combines the mathematical rigor of computational biology with the reasoning capabilities of large language models.

Rather than asking AI to interpret raw biological data on its own, the platform first organizes complex single-cell and spatial transcriptomic information into a structured framework that AI can analyze more accurately and consistently.

The result is a system that helps researchers move beyond identifying which cells are present in diseased tissue to understanding how those cells communicate with one another — and which conversations may be driving disease.

Cell-to-cell communication governs virtually every aspect of human biology, from tissue repair to immune responses. When those signaling pathways become disrupted, diseases such as Alzheimer's disease and cancer can emerge. Yet mapping these interactions has remained one of biology's most computationally challenging problems.

"The biggest challenge in modern biology isn't generating data — it's interpreting it," says Dr. Wong, the John S. Dunn Presidential Distinguished Chair in Biomedical Engineering at Houston Methodist. "We designed iS2C2 to bridge that gap by combining rigorous computational analysis with AI that can explain complex biological mechanisms in ways researchers can immediately investigate."

Unlike conventional AI approaches that may hallucinate or generate unsupported conclusions, iS2C2 is built around a transparent workflow. The platform first uses newly developed computational algorithms to identify biologically meaningful signaling pathways from single-cell RNA sequencing and spatial transcriptomic datasets. Those findings are then presented to a large language model through carefully structured prompts that encourage evidence-based reasoning rather than speculation.

The investigators evaluated the platform using datasets from Alzheimer's disease and cancer. In Alzheimer's disease, iS2C2 identified previously underrecognized communication pathways between neurons and neighboring support cells that may contribute to neurodegeneration. In cancer, the platform uncovered signaling networks involved in bone metastasis and highlighted a potential combination treatment strategy that could interrupt those pathways earlier in disease progression.

Just as importantly, the researchers demonstrated that the platform remained accurate even when working with incomplete sequencing datasets — a common limitation in real-world single-cell research. By incorporating generative AI to recover missing biological information, iS2C2 maintained robust performance while producing findings that domain experts judged to be reproducible and biologically interpretable.

For Dr. Wong, the significance extends well beyond any single disease.

“Understanding disease ultimately comes down to understanding how cells communicate and where those conversations break down. If we can identify which cells are driving disease, how they’re communicating and which pathways can be interrupted therapeutically, we can create a far more actionable roadmap for precision medicine.”

---

Stephen Wong, PhD

As single-cell technologies continue producing increasingly complex datasets, platforms like iS2C2 could help researchers spend less time deciphering massive volumes of biological information and more time testing new hypotheses.